Re: a fast bloat measurement tool (was Re: Measuring relation free space)

Abhijit Menon-Sen <ams@2ndquadrant.com>

From: Abhijit Menon-Sen <ams@2ndQuadrant.com>

To: Andres Freund <andres@anarazel.de>

Cc: Amit Kapila <amit.kapila16@gmail.com>, pgsql-hackers <pgsql-hackers@postgresql.org>, Tomas Vondra <tomas.vondra@2ndquadrant.com>, Simon Riggs <simon@2ndquadrant.com>

Date: 2015-05-11T11:27:08Z

Lists: pgsql-hackers

Attachments

0001-Make-pgstattuple-use-heap_form_tuple-instead-of-Buil.patch (text/x-diff) patch 0001
0002-Add-pgstatbloat-to-pgstattuple.patch (text/x-diff) patch 0002

Hi Andres.

I've attached an updated patch for pgstatbloat, as well as a patch to
replace two uses of BuildTupleFromCStrings() elsewhere in pgstattuple.

I've made the changes you mentioned in your earlier mail, except that I
have not changed the name pending further suggestions about what would
be the best name.

Also:

At 2015-05-09 15:36:49 +0530, ams@2ndQuadrant.com wrote:
>
> At 2015-05-09 02:20:51 +0200, andres@anarazel.de wrote:
> >
> > I haven't checked, but I'm not sure that it's safe/meaningful to
> > call PageGetHeapFreeSpace() on a new page.
> 
> OK, I'll check and fix if necessary.

You're right, PageGetHeapFreeSpace() isn't safe on a new page. I've
added a guard to that call in the attached patch, but I'm not sure
that's the right thing to do.

Should I copy the orphaned new-page handling from lazy_scan_heap? What
about for empty pages? Neither feels like a very good thing to do, but
then neither does skipping the new page silently. Should I count the
space it would have free if it were initialised, but leave the page
alone for VACUUM to deal with? Or just leave it as it is?

-- Abhijit